From: Andreas Ericsson <ae@op5.se>
Date: Wed, 13 Feb 2008 13:58:51 GMT
Subject: Re: [PATCH 1/2] config api: Add git_config_magic_int()
Message-ID: <47B2F79B.9090406@op5.se>
In-Reply-To: <7vir0t3l9s.fsf@gitster.siamese.dyndns.org>

Junio C Hamano wrote:
> Andreas Ericsson <ae@op5.se> writes:
> 
>> +int git_config_magic_int(const char *name, const char *value,
>> +			 const char *magic_value, int magic_setting)
>> +{
>> +	if (value && !strcasecmp(value, magic_value))
>> +		return magic_setting;
>> +
>> +	return git_config_int(name, value);
>> +}
> 
> I do not think this has much to do with any "magic".
> 
> An instruction "use 0 threads" when taken literally would mean
> "do not use any CPU" which would not make much sense.  In that
> sense, giving a magic meaning of "guess an appropriate value" to
> 0 may be a good idea.  A valid alternative would be to make 0
> mean the same thing as 1, but that is much more boring ;-)
> 
> But if you did so, that means "var = 0" invokes the same magic
> as "var = auto".  The magic lives in "0", and not in "auto".
> 
> I think the direction your patch leads us is good, but I think
> it should allow an array of symbolic ways to spell values to be
> useful, that is:
> 
> 	struct config_symbolic_int {
>         	const char *name;
>                 int value;
> 	};
>         int git_config_symbolic_int(const char *var, const char *value,
> 				    struct config_symbolic_int *);
> 
> That way, you can have
> 
> 	{ { "high", 9 }, { "default", 0 }, { "low", 1 } };
> 
> and say things like "zlevel = high|default|low".

Good idea. I'll look into it right away. I sorely need a break from
coding backends for webapps anyway. I'll send this one separately and
then tack 2/2 onto the builtin-pack-objects patch thing. That one
should probably be a /3 series anyway, when I come to think of it.

-- 
Andreas Ericsson                   andreas.ericsson@op5.se
OP5 AB                             www.op5.se
Tel: +46 8-230225                  Fax: +46 8-230231